-
Bryan Lim, Sercan Arik, Nicolas Loeff and Tomas Pfister. Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting. https://arxiv.org/pdf/1912.09363.pdf
-
A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, I. Polosukhin, Attention is all you need, in: NIPS, 2017.
-
S. S. Rangapuram, et al., Deep state space models for time series forecasting, in: NIPS, 2018.
-
S. Li, et al., Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting, in: NeurIPS, 2019.
The focus of this project is to better understand how we can deal with high-dimensional time series with multiple inputs, missing values and irregular timestamps. In this situation, the performance of the classical approaches is not satisfying, and naïve applications of deep learning also fail. Even the deep learning models that have shown promising results tend to "black boxes" with little insight into how to interpret the results.
- Discover interplay between classical approaches of time series prediction and modern deep learning techniques.
- Reproduce the model in paper "Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting" by Bryan Lim, Sercan Arik, Nicolas Loeff and Tomas Pfister.
- Test the model on generated data and compare my results on datasets used by the authors in the paper.